Simple "Homemade" Sensors Solve Tough EMI Problems
Electromagnetic interference (EMI) and its suppression are major issues in most current system designs. In Europe, essentially no electrical product can be sold without the CE mark—certification that it is emission-free. Troubleshooting a "leaky" design can be difficult and expensive, however. Fortunately, designers can build simple electromagnetic sensors that will detect emissions from the gigahertz range down to audio frequencies. These sensors are simple enough to be constructed by anyone seriously interested, yet effective enough to solve serious design problems.
An anecdote about how these sensors solved a near career-limiting EMI problem is a good place to start. It illustrates some fundamental principles of RF-leakage detection and leads into a description of how to make practical RF sensors.
To start a serious story with a little humor, what really solved the problem illustrates again the old engineering saying, "Even a blind pig gets an acorn now and then—if he keeps rooting around." As you'll see, rooting around was an essential element of the problem solution. The sensor described below has been used by the author and others since 1973. The concept is presently applied in circuit design environments, as well as under the pressures of a stopped production line.
The serious EMI problem that started this discussion occurred in a Signal Corps 8-GHz transmitter then completing its development. It was about done and the qualification unit was in final test. Then it failed to meet radiated emissions limits. As I recall, it was a 300-channel unit for analog voice multiplexing. The government specs were tight, as usual.
The unit had passed all of its functional requirements, such as noise-power ratio, stability, power output, noise, distortion, reliability, temperature extremes, humidity, and so on. Failing the EMI requirements at the last minute was a crisis. Delivery was due or overdue. It's hard to imagine a more stressful situation, especially when the problem is new, unexpected, and unusual. Failure to make deliveries is serious business.
The transmitter's internal construction was similar to most microwave equipment of the time. It had several rack-mounted RF subassemblies that were plumbed with miniature rigid coax connections. In this case, the vertical racks had been replaced by a 6-ft. tall, "EMI-tight," RF-gasketed cabinet with a swinging front door. The oscillator, multipliers, load isolators, amplifier equipment, etc., were all in the top 25% of the cabinet. The space below was completely empty. It was a 4-ft. tall, open unoccupied space.
For EMI test, the cabinet was placed in an 18- by 24-ft. screen room containing all the sensing and signal-generating equipment to measure both RF radiation and susceptibility. With the 8-GHz power amplifier operating, an escaped 8-GHz signal was detected outside and inside the closed cabinet door. Of course, that signal was found at greater intensity inside the cabinet. Recall that all of the RF-generating capability was in the equipment in the top 2 ft. of the cabinet. There should have been no leaking RF in those top 2 ft.
I should mention here that in screen rooms capable of meeting mil-spec requirements, there's a standard family of generators, receivers, sensors, and antennas. For the 8-GHz signal, a square trapezoidal horn is used as an antenna. It's about 7 by 7 in. at the open end. Inside this horn, at the small end of the trapezoid, a tiny loop antenna couples signals from the flared opening to a 50-Ω coaxial cable. That flexible cable is connected to a tunable RF voltmeter, scope, or other display, such as a spectrum analyzer. Thus, any signal received by the horn antenna is captured from its approximately 50-in.2 opening (7 by 7 in.) and concentrated to the small loop feeding the coax.
Reported symptoms were that the highest RF levels of leakage were at 8 GHz and at their maximum in the lower left rear of the cabinet—the empty space! Escaping radiation could not be traced to its source. Due to schedule urgency, I immediately got heavily involved. I personally held the trapezoidal antenna and moved it to every possible location trying to find a maximum signal level that would indicate some point of signal origination.
It didn't take long to realize that we were searching with wide-angle field glasses (no pun) where a microscope would be appropriate. With the standard mil-spec measuring horn sensor we had adequate sensitivity, but little spatial resolution. The size of the 7-in. horn prevented it from getting into local discrete areas of potential leakage. Clearly, what was needed was a sensor or pickup probe with more resolution, even if some sensitivity had to be sacrificed. After some discussions with the project engineer, I went off to try an idea that didn't engender much support when I first mentioned its concept.
There being no suitable RF sensors available, I went home that night and wound about 19 turns of 41-AWG enameled copper wire around the end of a shortened, round toothpick. I terminated the two end leads in a BNC connector and epoxied the assembly to a short length of 0.090-in.-diameter brass tube. The tube was soldered to the rear of a BNC connector (Fig. 1). The sensing coil, in diameter, was about 0.050 in.
If handled with care, the assembly was just strong enough. I had no idea what RF pickup characteristics this sensor would have. It was just something small and easy to try. Realizing it was a shot in the dark, I immediately proceeded with another parallel but somewhat different approach as a backup. I thought this second probe was likely to be useful at lower frequencies. That proved to be correct.
To build a sensor with more sensitivity than the toothpick coil, and that's still small enough to allow probing in tiny places, I scouted up some small, unidentified, ferrite "donut" cores. I ground one in half. It was a little over 1/8 in. in outer diameter.
I did most of the grinding on my tool grinder, which has a wheel for carbide. I followed that with lapping on silicon-carbide "wet or dry" paper. This gave me a half-circle core of ferrite with two coplanar flat surfaces. This is a delicate operation in which fingers get clumsy. Several cores flew off the wheel before I successfully held one to completion. Figure 2 shows the approximate shape.
After a few failures, I discovered a way to assemble this ferrite-core probe with its many turns of 41-AWG wire. I wound turns until they approached the level of the two coplanar flats. Each end of the coil was soldered to 20-AWG lead extensions. They, in turn, were soldered and epoxied to a small brass tube which was soldered and epoxied to the BNC connector. Enamel coating was dissolved off the wires where they were to be soldered. I still had no idea what frequency response or sensitivity either probe would have. The construction is easier to illustrate than to explain (Fig. 3).
As soon as the toothpick sensor was available, the project engineer and I took it to the radiating unit. We tuned the receiver to the transmitter's 8-GHz frequency and set the gain at maximum. We probed the toothpick sniffer over "everything" with no indicated pickup. Knowing that the probe had a low sensitivity, it was a painstaking effort to trace around every possible location for the leak. In the most likely places, there was none. We didn't know how close you would have to be to sense any leakage.
The first candidates we checked were all the micro coax connections. Then, we surveyed the oscillator and multipliers, also with no results. Covers and joints were dutifully sniffed. It wasn't hard at that point to believe that this scheme was going to be a failure. But rooting around continued until—voilà—a response!
That response was at a location where it should've been impossible to have a leak, however. Was something wrong with the concept or implementation?
The RF maximum occurred in a tiny location on the side of a solid metal tube extending from a load isolator to its output connector. The isolator is a bit smaller than a box of kitchen matches. It has a rigid input coax at one end and a similar rigid coax at the other end. The leak was centered on the tube extension's side, where unblemished gray paint showed no clue (Fig. 4).
As it turned out, the load-isolator manufacturer had drilled a small manufacturing access hole in the side of the metal tube. The arrow pointing at the tube shows its location (Fig. 4, again). Using the hole to gain access, the company could solder the center conductor of the coax at that point. It dutifully filled the hole with epoxy, smoothed it, and then gray-painted the entire assembly. Masking kept paint from an adjacent silver-plated area for the mating connector—an innocent act having diabolical results.
We were so glad to have found the problem that I don't recall giving the isolator vendor a hard time. The leaking hole was covered with metal and the problem was solved. The larger horn antenna was used in the official test, which showed that the RF leak had been eliminated.
With the problem out of the way, immediate trials of the second pickup design were deferred. As anticipated, the ferrite-core design proved to have greater sensitivity. It had an amazingly flat bandwidth. I wish I'd had the time to characterize the probe starting with known ferrite, turns count, etc. It would be interesting to know the bandwidth, sensitivity, and other characteristics. But that proved impossible at the time. In the intervening years, I've done a little characterization—just enough to be sure of getting good results. Unfortunately, that data is gone. I'll talk about additional work in a moment.
This success kicked off further thinking about the principles involved and how to maximize signal sensitivity while minimizing the probe size. In later models, the ferrite core proved to be essential. It improves sensitivity and extends response to low frequencies.
Figure 5 explains why the mil-spec antenna failed in this troubleshooting task, and how a simple homemade sensor pointed to a solution. This diagram shows an RF-tight enclosure containing a signal source. A small hole in the top is the point of RF leakage to the outside. A quick reference to field theory shows that the leaking E-field and H-field lines in this cross section are at right angles everywhere.
Considering this in 3D, it's easy to see that if you captured all the energy in the area marked A, it all had to come from and be present at the smaller area, B. The larger horn pickup that couldn't resolve location of the leakage was receiving the same total energy that had emerged at a higher density closer to the hole. Logically, a much smaller probe with lower sensitivity could still sense the same energy, but only at closer distances from the leak. It's simply a tradeoff between resolution and sensitivity.
It should be obvious that the ferrite probe has many additional uses. It's worked out well for use in equipment design and test. At lower frequencies, the opportunities multiply. Later probes, built with cores up to about 3/16-in. outside diameter, helped trace signals on connector wires and on circuit-board traces. You can even define current paths in sheet conductors like ground planes. The emergence of network analyzers, high-frequency scopes, and spectrum analyzers makes this probe a potent design tool. It's also great for digital signal tracing.
Personally, I've built about 13 or 14 of these probes, most of which were left with the engineers I was helping. They've been used in avionics, missile-guidance systems, audio, analog multiplexing, FM amplifiers, and most recently in applications on the International Space Station dc-dc power converters. That's still going on as this is being written.
Since this idea was conceived in 1973, a few other probes have come on the market. They're fine for their intended purpose. But with this design, you get a home-built probe in use before you can get a requisition through purchasing.
I no longer have access to up-to-date test equipment. So the notion of a full characterization is no longer possible. Every time I built a probe for a specific application, however, it worked sufficiently to kill the problem. The range of expected frequencies determined the core size and turns count. We did check the ferrite characteristics for the choice of core. But we never were sure of just how much margin there was. If one of you readers chooses to try this probe concept, be prepared to be amazed at the extremely flat frequency response and broad bandwidth. Note particularly the absence of bothersome resonances. Perhaps someone with adequate equipment will pick up where I left off and quantify all these variables.
This article has emphasized higher-frequency applications. I built a couple of larger pickups wound on 1/4-in. cores. They provided good pickup down into the audio area. I have traced 9-kHz signals through wires, boards, racks, and chassis with one of the larger coils.
Figure 6 shows the probes made for missile-guidance-system use that are currently being utilized on the International Space Station. The application's power converters handle 75 kW in switching supplies and have a lot of radiation potential at many frequencies. But the probes are extremely flat and wideband.
If an experimenter decides to go sniffing around with one of these probes, bear in mind that your hand is poking a conductor into areas of voltage. Be forewarned. Shrink sleeving around the exposed metal could help.
In use, one finds that rotating the probe resolves the direction of current in the sensed circuit. The probe is, in effect, a transformer in which the one-turn primary is the sniffed circuit. The probe coils are the secondary. The ferrite concentrates the flux. Orienting the probe for maximum pickup is a skill quickly and intuitively learned.